Abstract:To address the issues of low detection accuracy and high computational load in existing distracted driving detection algorithms, this paper proposes an improved model based on the YOLOv11n architecture, named YOLO-DCC. Firstly, the algorithm replaces the original YOLOv11n backbone with two distinct lightweight backbones, enhancing its feature extraction capabilities while maintaining a lightweight architecture. Subsequently, the Cross-Attention Fusion Modules (CAFM) are employed to achieve dynamic interaction and weighted fusion of cross-branch features along both channel and spatial dimensions, thereby enhancing the representation of key features. Finally, Context Guided Fusion Modules (CGFM) are introduced into the neck network to optimize the fusion of multi-scale features by integrating local, surrounding, and global contextual information, improving the model"s detection accuracy. Experiments on the StateFarm dataset show that the proposed algorithm achieves a mean Average Precision (mAP) of 95.7% at an IoU threshold of 0.5, a 3.1% improvement over the baseline YOLOv11n model, while reducing the computational load by 7.94%, demonstrating its effectiveness in detecting various distracted driving behaviors.